Hello everyone, I am the model maker Poakl.
This Lora is based on Flux Checkpoint and its text encoding to train. Its function is to effectively repair and improve the body and hands of Flux generated images when the weight is 0.3-0.6. At the same time, Lora has very little impact on other factors of the original image when used, and only optimizes and changes the body and hands.
We all know that Flux has made great progress in the body and hands, but in certain situations (such as yoga movements), there are still problems such as multiple hands and feet, and finger deformities. Therefore, I tested Lora 14 times, and after constantly adjusting the training set and parameters, I finally got this effective Lora.
How to use: No trigger words are needed, and after adding Lora, set its weight to 0.3-0.6.
Since the weight ratio of Lora is different in different operating environments, after testing on tensor raw images, I got the following usage suggestions:
If you only want to use Lora to repair the hands and body, please set the weight to 0.2, 0.3 or 0.4, which will play a repair role while affecting the original image to the minimum extent.
If you want to make major modifications and repairs on the basic composition of the original image, please set the weight between 0.5-0.6. A larger weight will make the image too white.
We are still in the stage of exploration in the training of Flux. I hope everyone will use my Lora more, and welcome more pictures and comments~!
